C OMPOSITIO M ATHEMATICA
C. J. A TKIN J. G RABOWSKI
Homomorphisms of the Lie algebras associated with a symplectic manifold
Compositio Mathematica, tome 76, no3 (1990), p. 315-349
<http://www.numdam.org/item?id=CM_1990__76_3_315_0>
© Foundation Compositio Mathematica, 1990, tous droits réservés.
L’accès aux archives de la revue « Compositio Mathematica » (http:
//http://www.compositio.nl/) implique l’accord avec les conditions gé- nérales d’utilisation (http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une in- fraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
Homomorphisms
of the Liealgebras
associated with asymplectic
manifold
C. J. ATKIN and J. GRABOWSKI
1Department of Mathematics, Victoria University of Wellington, P.O. Box 600, Wellington, New Zealand, 2J. Grabowski, Instytut Matematyki U.W., P.K. i N. IXp. 00-901 Warszawa, Poland
© 1990 Kluwer Academic Publishers. Printed in the Netherlands.
Received 5 November 1988; accepted in revised form 30 October 1989
1. Introduction
We have each given (in [1] and [7]) proofs of an algebraic result - restated for
our present purposes as (6.2) below - which in effect (see (6.5)) constructs a one- one correspondence between the points of a symplectic manifold and certain
subalgebras of its Lie algebra of Poisson brackets. Our aim here is, firstly, to
extend this result to the Lie algebras of locally, globally, and conformally
Hamiltonian vector fields determined by the symplectic structure; and then to utilise it to prove that each of these algebras determines the manifold, as far as
that is possible. In fact, we approach these ’uniqueness theorems’ by studying
certain types of Lie homomorphism (which we classify in Section 7) between
such Lie algebras, rather than by reconstructing the manifold from the algebra;
this method (modelled on that in [6]) both gives and requires less structural
information, but yields more facts about homomorphisms. Indeed, we have
taken no pains to delve more deeply into the structure of our algebras than our techniques demand, and those techniques are perhaps more interesting than the
results which motivated them.
In Section 2 we present some definitions and facts not related to symplectic
structures; Section 3 introduces the notion of n-ample algebras of vector fields.
In Sections 4 and 5 we review some definitions and notations, and give proofs of
some properties which will be needed subsequently (and one or two which will
not, such as (5.8)). Here we mostly follow, and often refer to, the well-known paper [2]. Although we allow both the real-analytic and the holomorphic (Stein) differentiability classes, the results from [2] which we quote are merely local,
and as such hold in these cases without any modification. Then Section 6 gives
the algebraic characterisations of the points of a symplectic manifold, Section 7
classifies suitable homomorphisms, and Section 8 considers the application to epimorphisms and isomorphisms.
Special cases of the ’uniqueness theorem’ have been proved before (though
not, we believe, published). Our method, however, seems to be the first which
applies simultaneously to so many cases.
2. Preliminaries on manifolds and on Lie algebras
(2.1 ) In speaking of an n-dimensional manifold M of dass W over the field F, we shall mean one of three things:
(a) a real paracompact manifold of differentiability class Coo and of real
dimension n, when W denotes Coo and F denotes the real field R;
(b) a real paracompact manifold of differentiability class C03C9 and of real dimension n (when W denotes C03C9 and F denotes R);
(c) a complex manifold of complex dimension n for which each connected component is Stein (when W denotes the holomorphic differentiability class
Jf and F denotes the complex field C).
We shall not consider complex manifolds whose components are not Stein,
and shall often omit explicit mention of Y, F, or n.
(2.2) For each of (a), (b), (c), one has an embedding theorem (due to Whitney [ 14] in case (a), to Remmert and to Narasimhan [12] in case (c), and to Grauert [8] in case (b)): a connected manifold of dass W and dimension n is CC-
diffeomorphic to a closed Y-submanifold of Fln + 1. (Note that by a ’closed
submanifold’ we understand a ’properly and regularly embedded submanifold’.) (2.3) For a manifold M of class W, we denote by TM the bundle of tangent
vectors (meaning, in case (c), the tangent vectors of type (1, )), and by T*M the corresponding cotangent bundle. The vector space over F of exterior forms of dass W on M (again, in case (c), these forms are to be of type (k, 0) and holomorphic) will be called
S2k(M).
The exterior derivatived:03A9k(M) ~ 03A9k+1(M)
is defined as usual; we write its kernel as
Z’(M)
and its image as Bk+ l(M), withthe convention that
03A9k(M)
= 0 when k 0. The chain complex(03A9k(M),
d) is thede Rham complex of M.
(2.4) LEMMA. Let M be a manifold of class W; let p, q E M, p ~ q, and let k be a
nonnegative integer. Given any k-jets (of F-valued functions) at p and q, there
exists f
en’(M)
which has those k-jets.Proof. When M = F", this is trivial. (2.2) then gives it in general.
(2.5) For each of the cases of (2.1), there is a de Rham isomorphism
where the cohomology may conveniently be assumed singular. For case (a), this
is de Rham’s theorem. For case (c), it is a well-known consequence of Cartan’s Theorem B; see, for instance, [4], exposé XX, or p. 80 of [9]. The same argument
may be applied in case (b), where Tognoli [13] has pointed out the validity of
Theorems A and B in a real-analytic version.
(2.6) PROPOSITION. Let M be a manifold of class CC and dimension n; take
m = 2n + 1. Then there exist F-valued functions x1,..., Xm of class CC on M, such that, for any integer k ~ 0 and any form 03C8 E
0’(M),
there exist F-valued functions fi1,i2,....ik of class W on M ( for all i1, i2, ... , ik with 1il i2 ... ik ~ m), forwhich
Proof. It will clearly suffice to prove the result for each component of M individually. So we may suppose (see (2.2)) that M is Y-embedded in Fm, with embedding j : M ~ Fm. The natural monomorphism TM ~ j*TFm defined by Tj
dualises to epimorphisms 03C0~: A
~T*Fm|j(M) ~
A tT* M for any ?. Let!)k(M),
!)k(Fm)
denote the sheaves of germs of k-forms of class on M, Fm respectively;then nk induces a sheaf epimorphism (over M)
Let Q = ker J, so that there is an exact sequence
Let Cm be the structure sheaf over M of germs of COO functions in case (a), C03C9
functions in case (b), and holomorphic functions in case (c). Then J is a homomorphism of (9,-modules, so that Q is also an (9,-module.
Let Y1, ..., y. be the coordinate functions on Fm. Then
fl’(F’)
is free over the appropriate structure sheaf (9F-; indeed, it has free generators given by thesections dyi ^ ··· A dytk for 1 ~ i ... ik ~ m. Hence, as j*OFm = (9, trivi- ally,
j*03A9k(Fm)
is free over (9m, and it has free generators (dyi ^ ··· A dyi,)-induced from the sections
In the C03C9 and holomorphic cases
!)k(M)
is coherent over OM, and so of courseis the free OM-module
j*03A9k(Fm).
By Serre’s 3-lemma, then, Q is also coherentover (9m. From Theorem B (see (2.5)) we know that
Hk(Q)
= 0 for k > 1.In the COO case, (9m is soft (see [3] or [5]) so that Q is also soft and
H1(Q)
= 0.In all three cases, the cohomology exact sequence of (1)
leads to the result that J* is onto. The given form 03C8 e
03A9k(M)
is consequently theimage J* of a section 0
ofj*fik(Fm);
but, as remarked at(2),j*!)k(Fm)
is free over (9m, so that there aresections fi1...ik
of (9, for whichApplying J* (that is, compounding with J), we obtain after some bookkeeping
where, for each i, xi = Yi 0 j; this is clearly the result.
(2.7) Again let M be a manifold of class Y. Then r(M) will denote the Lie algebra
of sections of TM of class Y. If 5F is a foliation of M of dass W, let 0393(F) be the
Lie subalgebra of r(M) consisting of vector fields everywhere tangent to 97. In general, if K is a vector subspace of r(M) and p E M, set
(2.8) If oc E
S2k(M),
let ro(a) denote the class of vector fields of class which leaveoc invariant, and let r(a) be the class of vector fields which operate on a as multiplication by a locally constant function. In other words, if £f denotes the Lie derivative,
As Y[X,Y] = [YX, 2 y], it follows that
so that 0393(03B1) is a Lie subalgebra of r(M) and 03930(03B1) a Lie ideal in r(a) (including its commutator).
(2.9) For any Lie algebra L over F, let Z(L) denote the class of all self-
normalising maximal proper finite-codimensional Lie subalgebras of L. (Notice
that a maximal subalgebra is self-normalising if and only if it is not an ideal, and
that it can be an ideal if and only if it includes the commutator.)
We may call 1:(L) the ’spectrum’ of L.
Let L(n) denote the nth. derived ideal of L, for n = 0, 1, 2, ... ; thus L(0) = L and,
for each n, L(n+1) = [L (n),
L(n)].
It will be convenient for technical reasons todefine the ’n-spectrum’ E"(L) (for n = 1, 2, 3, ... ) as the class of subalgebras Q of
L such that Q E 03A3(L) and also
We have already observed that
Since L(n) is an ideal in L, Q + L(n) is a subalgebra; thus for Q E 1:(L), (1) is equivalent to
(2.10) LEMMA. Let 4): L1 ~ L2 be a surjective homomorphism of Lie algebras
over F. Then, for any positive integer n,
Proof. Certainly
03A6(q’), 03A6-1(Q)
are finite-codimensional Lie subalgebras of L2, L 1 respectively. Let R be a Lie subalgebra of L2 such that R ;:2 03A6(Q’).Then 03A6-1(R) ~ Q’;
consequently,either 03A6-1(R) = Q’ or 03A6-1(R) = L1,
and,as 03A6 is surjective, R =
03A6(03A6-1(R))
is either O(Q’) or O(Li) = L2. Hence O(Q’)is either L2 or a maximal proper Lie subalgebra. Similarly, let S be a Lie subalgebra of L1 such that S ~
03A6-1(Q);
then as 8;:2(D - ’(0), S = 03A6-1(03A6(S)).
But 03A6(S) ~ Q. Therefore 0(5) = Q or 0(5) = L2, and either S
= 03A6-1(Q)
orS = Li;
hence 03A6-1(Q)
is a maximal proper Lie subalgebra of L1.Now, if
03A6-1(Q)
were a proper Lie ideal of L1, Q =03A6(03A6-1(Q))
would be aproper Lie ideal of L2, since 03A6 is surjective; if 03A6(Q’) were a proper Lie ideal of L2,
03A6-1(03A6(Q"))
would be a proper Lie ideal in L1 including Q’, and therefore wouldbe equal to Q’. But neither Q nor Q’ is a Lie ideal (in L2, L1 respectively); so
03A6-1(Q)
and O(Q’) are not Lie ideals. This proves (a) and (b) when n = 1.Finally, suppose n > 1. Then, as 03A6 is epimorphic,
whilst, if
03A6-1(Q) ~ Lln),
then Q =03A6(03A6-1(Q)) ~ 03A6(L(n)1)
=L(n)2.
The results nowfollow, by (2.9)(3) and (2.9)(1) respectively.
(2.11) LEMMA. Let L be a Lie algebra over F, and let K be a Lie ideal of L including L(1).
Then
(a) if (L, K] = K(1) and Q E 1:(L), there exists NeE(K) such that Q n K z N;
(b) if, for some positive integer n, Q E yn 1
l(L),
there exists N E 03A3n(K) such that Q n K z N.Proof. Consider case (a). As K is an ideal, K + Q is a Lie subalgebra of L.
However, K e Q; for otherwise, as K ~ L(1), we should have Q ;:2 L(1) and Q
would be an ideal. Hence K + Q ~ Q and, by maximality, K + Q = L.
By hypothesis, [Q, K] ~ [L, K] = K(’). Thus, if K(1) ~ Q, K must normalise Q and (as K + Q = L) Q must be an ideal in L. This is false, as Q E Z(L); we
deduce that
In case (b), K(n) ~ L(n+1) and, as Q E
03A3n+1(L),
it follows immediately thatIn either case, K(1) + Q is a subalgebra (as K(1) is an ideal) which is not equal
to Q; thus K(’) + Q = L, and as a consequence
As K ~ Q, K n Q is of finite positive codimension in K. Let N be a maximal proper subalgebra of K including Q n K (which we may construct by finite induction). Then, by (2), K(’) + N = K; in view of the maximality of N, this implies that N ";/2 K(1) and N is not a Lie ideal in K (see (2.9)). This proves (a). For (b), observe that (1) gives
exactly as (0) led to (2). Ergo, K(") + N = K, which shows N E En(K), by (2.9)(3).
3. Lie algebras of vector fields
(3.1) Once more, let M be a manifold of class W and let L be a Lie subalgebra (over F) of r(M) (see (2.7)). We shall say that L is n-ample, where n is a positive integer - more precisely, L is an n-ample subalgebra of r(M) - if, for each p E M, Lp E 03A3n(L) (see (2.9)).
(3.2) LEMMA. (a) Let L be an n-ample subalgebra of r(M). Then
(b) Suppose L is a 1-ample subalgebra of r(M) and satisfies (1). Then L is n- ample.
Proof. (a) By hypothesis, Lp + L(") = L (see (2.9)(3)). The result follows, as
L,(p)
= 0 by definition.(b) If L(n)(p) = L(p), then evidently Lp + L(n) = L. Apply (2.9)(3).
(3.3) LEMMA. Suppose L1, L2 are Lie subalgebras of r(M) such that L1 £; L2 and, for any p ~ M, L1(p) = L2(p). Then, if L1 is n-ample, so is L2.
Proof. Take p E M, and let R be a Lie subalgebra of L2 which includes
(L2)p.
Then R n L1 ;:2
(L1)p,
and, as L1 is 1-ample, either R n L1 = L, orR n L1 =
(L1)p.
As L1( p) = L2(p), certainlyConsequently
If R ~ L1 = Li, (2) and (1) show that R = L2 ; whilst, if R ~ L1 =
(L1)p,
(2) showsthat R =
(L2)p.
This establishes the maximality of the subalgebra(L2)p
of L2,and it is evidently of finite codimension therein. If it were an ideal in L2,
(L1)p
= Li n(L2)p
would be an ideal in Ll, which it is not. This shows that, if L1is 1-ample, so is L2. If Li is n-ample, by (3.2)(a)
so that L2 is n-ample by (3.2)(b).
4. Symplectic structures and the associated Lie algebras
(4.1) A symplectic manifold (M, co) of class W and dimension 2n is a manifold M of class and dimension 2n, furnished with an everywhere non-degenerate
closed 2-form 03C9 E
Z2(M).
Following [2], p. 2, we have then a bundle isomorph-ism of class Y,
(where i denotes the internal product), which induces isomorphisms, also
denoted by 03BC03C9, of the tensor bundles and their spaces of sections.
(4.2) We have also A = Aro =
i(03BC-103C9(03C9)):03A9r+2(M) ~
gr(M) (ibid., p. 3).(4.3) Let L*(co) = y,
1(B1(M))
and L(co) =03BC-103C9(Z1(M))
denote respectively thespaces of globally and of locally Hamiltonian vector fields on M. Certainly
so that L(03C9) and L*(cv) are Lie subalgebras of the algebra of vector fields, and (recall (2.8)) [0393(03C9), 0393(03C9)] ~ Iro(co) = L(cv) «3. 1) on p. 6 of [2]). Following ([2], p.
11), we describe fields in 0393(03C9) as ’conformally Hamiltonian’.
(4.4) A foliation F (of class W) of the symplectic manifold (M, w) of dass W will itself be described as ’symplectic’ if w restricts to an everywhere non-degenerate
form on each leaf of ,9’. Thus the leaves also become symplectic manifolds.
Similarly, a subbundle S (of class Y) of TM is ’symplectic’ if cv restricts to a non- degenerate form on each fibre of S. Clearly there is the usual correspondence
between symplectic foliations and integrable symplectic subbundles.
(4.5) Given f, g E
QO(M),
one defines the Poisson bracket (relative to w) byby an easy computation. This makes
Q°(M)
into a Lie algebra, which we denote by A(M). Using the non-degeneracy of w and (2.4), one sees that the centre C(M)of A(M) consists of the locally constant functions of dass W on M; that is,
Thus d induces a linear isomorphism A(M)/C(M) ~
B1(M),
which determinesa Lie algebra structure on
B1(M);
then 03BC03C9: L*(03C9) -B 1 (M)
is a Lie algebra isomorphism. There is a Lie algebra exact sequence(4.6) For X, Y ~ L(03C9), one has ([2], (3.3), p. 7)
(4.7) The Lie algebra A(M) is also a commutative associative algebra under pointwise multiplication of functions, which is related to the Lie algebra
structure by the structural equation derived from (4.5)(1)
(4.8) PROPOSITION. For every n ~ 1 and every p E M,
Proof. To avoid messy calculations, let us use the theorem of Darboux to construct coordinates (xl, ... x2n) of class W on a neighbourhood of p such that
03C9 is represented on that neighbourhood as
03A3ni=1 dxi
A dXn+i’ In these coor-dinates,
y. 1
has the formthus, in terms of principal parts,
03BC-103C9
is represented by a constant linear isomorphismFor a scalar-valued function f, and x in the chart in question, df(x) is represented by the derivative Df(x) E
(F2n)*;
thus J(Df(x)) represents MI,1(df)(x),
and the kth.derivative in these coordinates of XI = 03BC-103C9
1(df)
at p is obtained by identifyingDk+1f(p) ~ (~k+1F2n)*
with a linear mapping ~kF2n ~(F2")*
and compoundingwith J.
(A) We now claim inductively that, given integers k >_ 0,0 ~ ~ ~ k, and nonzero
vectors 03BE ~ ~~ F2n, ~~F2n,
there existsXE(L*(w))(n)
such that D"X(p) = 0 when r ~ k and r ~ ~, whilstD~X(p). 03BE
= il. For n = 0, take X =X f,
where D1(p) = 0for r k + 1 and r ~ ~ + 1, and
D~+1f(p)
is a symmetric elementof (~~+1F2n)*
such that De +
1f(p) ·
(03BE Q r) =J-1(~) ·
i for each element T of a basis of F2". The existence of such a symmetric multilinear map is an elementary exercise using multinomials; the existence of a suitable f E A(M) follows from (2.4).Suppose the claim established for arbitrary k, ?, 03BE, ~ and given n ~ 0, and take
X E
(L*(03C9))(n)
so that X(p) :0 0, DX(p) = 0,D2X(p)
= 0,..., Dk+1X(p)
= 0. Thenfor any Y E r(M), the local coordinate representations of [X, Y] are
and so on to
Dk[X, Y](p) = Dk+1Y(p)·(,
X(p)) + terms combining lowerderivatives of Y and higher derivatives (to order k + 1) of X; by choice of X, then,
If Y is chosen in
(L*(w))(n)
so that DY(p)
= 0,...,Dty(p)
= 0,D~+2Y(p)
= 0, ...Dk+1 Y(p) = 0, and
D~+1Y(p) · (03BE
0 X( p)) tl, it follows that [X, Y] satisfies therequired conditions, and of course it is an element
of (L*(03C9))(n+1).
This proves the claim.Now take k = e = 0 and the Proposition follows.
Note. The result when M is real follows instantly from (5.5). In the complex
case, however, the above argument seems the simplest (though not the most conclusive) available, and is largely needed anyway for the next result.
(4.9) LEMMA. Let X E r(M), p E M, and X(p) ~ 0; suppose B is a subspace of r(M) which includes L*(w). Then the subspace of B
is of infinite codimension in B.
Proof. Use the coordinate system introduced in the proof of (4.8). Applying (4.8)(1) inductively, we find that in terms of these coordinates
+ terms in lower-order derivatives of Y at p;
but, by the assertion (4.8)(A), there exists for each n ~ 1 a field Y E L*(cv) such
that Y( p) = 0, D Y( p) = 0,..., D" Y( p) = 0, Dn+ 1 Y(p) · (X(p), X(p), ..., X(p)) ~ 0.
Thus
YjXY(p)
= 0 for 0 ~ j n, whilst YnXY(p) ~ 0. This evidently proves the result.5. Commutators in the Lie algebras
(5.1) In this section, (M, cv) is a fixed symplectic manifold of class and
dimension 2n, and we abbreviate the previous notations (from §§2, 4) A(M), Qr(M), zr(M), Br(M), J.lro’ Aro, [ , ]03C9, to A, 03A9r, Zr, Br, 03BC, , [ , ] respectively. As
in [2], p. 2, set ~ = wn/n! E Q2n, the symplectic volume form. We have
This follows
instantly
from L(cv) = 03930(03C9) (see (4.3)).The symplectic adjunction operator *, defined on p. 2 of [2], satisfies
For the symplectic coderivative £5: 03A9p ~ 03C9P-1: a ~ ( -1)p * d * a, one has
as is proved in [10] and quoted as (1.8)(a) on p. 3 of [2].
(5.2) THEOREM.
Moreover, there exist elements xl, X2, ..., Xm of A (where k = 4n + 1) such that
the mapping
is surjective.
Proof. Suppose first that g, h E A. Then, by (4.5)(2) and (5.1)(1),
(i denotes the interior product, and we have used the fact that Q2n =
Z2n).
Hence[g, h]~ ~ B2n.
For the converse, suppose f ~ A
and f~ ~ B2n.
So there exists03B2~03A92n-1
such that fri = d03B2, and consequentlyLet x1, x2, ... , xm be as in (2.6) (with the difference that the dimension of M is
now 2n). Hence, as
*/3 E QI,
there exist functions fI’ f2, ..., fm E A such thatIt follows that
by (5.1)(3), since A has degree - 2
by (4.5)(1). This shows that f E A(1), and completes the proof.
(5.3) COROLLARY. There is a canonical isomorphism A/A(1) ~
H 2n (M;
F).Proof. The map A --+ S22": f - fq is an isomorphism; by (5.2), it carries A(1) on
to B 2n . As 03A92n = z2n, the result follows.
(5.4) NOTES. Both A and
H2n(M;
F) are direct products of the correspondingfunctors of the individual components of M. When M is real (whether COO or cro),
1 will define the fundamental class of each compact component; for each noncompact component, the top cohomology vanishes. Thus the isomorphism
A ~ 03A92n:f ~ f~
of (5.3) induces, via inclusion C0 ~ A and projection 03A92n ~H2"(M;
F), an isomorphism of the space Co of locally constant F-valuedfunctions (COO or C03C9 as appropriate) of compact support on M with
H2n(M;
F).Since Co is clearly an abelian ideal of A, (5.3) now yields a Lie direct sum decomposition A = Co (3 A(1). As Co is central, one deduces in turn that A(1) = A(2).
These arguments do not hold in the complex case (2.1)(c). In that case, the top dimension for F-cohomology, namely 2n, is only the middle topological dimension, and, even for connected M, the dimension of H 2n(M; F) may be any finite integer or countably infinite. Indeed, let E be any discrete subset of C, and
endow M = (CBE) X
(CB{0})2n-1
with the trivial symplectic structure (as asubset of
C2n).
The 2n th Betti number of M is the number of points of E, and Mis clearly Stein.
Surprisingly few examples of compact real symplectic manifolds are known.
For a fairly recent, though inconclusive, survey, see [11]. There has been some
progress since (by Gromov and McDuff in particular).
(5.5) LEMMA. Let M be a real symplectic manifold, of class COO or C03C9. Then
Proof. By (4.5)(4), L*(03C9) ~ A/C(M); by (5.4), A =
C0 ~ A(1),
whereCo - C(M). Hence [L*(cv), L*(cv)] = L*(03C9). This suffices (see (4.3)).
(5.6) LEMMA. Let M be a real symplectic manifold of class COO or C03C9, and
suppose the symplectic form w is exact. Then
Furthermore, 0393(03C9) is the semi-direct product of its derived ideal L(co) with an
abelian subalgebra isomorphic to C(M).
Proof. See [2], p. 12. (We write r instead of L5, w in place of F; (5.5) must be
invoked in the CW case, and our formulation allows M not to be connected).
(5.7) LEMMA. Let M be a symplectic manifold of class W. Then
(Compare (5.5)).
Proof. In view of (4.5)(4), [L*(a», L*(cv)] consists of the images under
03BC-1d
ofelements of A(1), which are characterized by (5.2). Suppose XE L(w) and
Y E L*(cv), so that Y =
03BC-1(df)
for some f E A and !£ xw = !£ yw = 0 (see (4.3)).Hence, by (4.6),
and it is enough, by (5.2), to show that A(p(X) A df)~ E B2n.
Now
But
This completes the proof.
(5.8) LEMMA. Let M be a connected symplectic manifold of class CC, and suppose the symplectic form co is not exact. Then 0393(03C9) = L(cv) = 03930(03C9).
Proof. See p. 11 of [2] (after (5.4)); no change is needed.
Note that when M is compact, cv cannot be exact (as ri is not).
6. Spectra
(6.1) By a Poisson algebra over the field F, we mean a commutative associative
algebra A furnished with an F-bilinear operation
with respect to which it is also a Lie algebra over F, and satisfies the structural relation
for all f, g, h ~ A. Thus, in particular, A(M) is a Poisson algebra over F when (M, cv) is a symplectic manifold of class 16 (see (4.7)).
Given a Poisson algebra A over F, let m(A) denote the set of all maximal finite-codimensional proper associative ideals J in A, and let E(A) be as in (2.9).
The theorem which follows is proved (in superficially different formulations) in [7] and in [1].
(6.2) THEOREM. Suppose the Poisson algebra A satisfies:
(a) A2 = A,
(b) for any J ~ m(A), the Lie normaliser
is a proper finite-codimensional linear subspace of A.
Then the mapping J ~ 91 A(J) establishes a one-one correspondence between m(A) and E(A).
(6.3) Now let (M, 03C9) be a symplectic manifold of class CC and positive dimension.
Given p E M, define
On p. 17 of [6] it is proved that the map p H p* furnishes a bijection between M
and m(A(M)). In addition, (6.4) LEMMA.
Proof. Suppose g E A(M) and dg(p) = 0; then, by (4.5)(1), [f, g]03C9 ~ p*. This
proves that N(M, p) ~
91A(M)(P*)’
Conversely, supposeg ~ nA(M)(p*)
but dg( p) = 0. As 03C9 is non-degenerate at p, there exists XET*pM
for whichBy (2.4), there exists f E p* with df(p) = x. Thus [ f, g]03C9(P) ~ 0 and g e
nA(M)(P*).
This completes the proof.
(6.5) THEOREM. The map p H N(M, p) constitutes a bijection of M with 1:(A(M)).
Proof. As remarked in (6.3), p ~ p*: M - 03A3(A(M)) is bijective; hypothesis (6.2)(a) is trivial, and (6.2)(b) follows from (6.4). Thus the result follows from (6.2)
and (6.4).